In [1]:
from cntk.device import try_set_default_device, use_default_device, gpu
try_set_default_device(gpu(0))
use_default_device()
Out[1]:
In [2]:
import numpy as np
import sys
import os
import math
import time
import cv2
import matplotlib.pyplot as plt
ตัดเอกสาร¶
In [3]:
name = "NewData"
base_folder = "D:/cntk1.1/Examples/Image/DataSets/MNIST/OCR-Database/"+name #เปลี่ยนชื่อเอา
files = next(os.walk(base_folder))
file_list = files[2]
for i in range(len(file_list)):
#if '.png' or '.PNG' or '.jpg' in f:
image_path = base_folder + "/" + files[2][i]
#print(image_path)
img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
thresh_value = 175
max_value = 255
thresh_value, thr_img = cv2.threshold(img, thresh_value, max_value, cv2.THRESH_BINARY)
thr_img = max_value - thr_img
cv2.imwrite(base_folder+"/threshold.png", thr_img)
connectivity = 8
labeling_outputs = cv2.connectedComponentsWithStats(thr_img, connectivity, cv2.CV_8UC1)
# main data elements for connected component extraction
num_labels = labeling_outputs[0]
labels = labeling_outputs[1]
stats = labeling_outputs[2]
centroids = labeling_outputs[3]
rows = img.shape[0]
cols = img.shape[1]
output_folder = base_folder + "/" + name +"/"
#output_folder2 = "C:/xampp/htdocs/DeepOCR/" + name +"/"
for k in range(len(file_list)):
#output_folder = output_folder +"/"+ str(k) + "_"
try:
os.mkdir(output_folder)
#os.mkdir(output_folder2)
except OSError as exc:
print("folder " + output_folder + " exists")
#print("folder " + output_folder2 + " exists")
j=0
for i in range(0, num_labels):
left = stats[i, cv2.CC_STAT_LEFT]
top = stats[i, cv2.CC_STAT_TOP]
height = stats[i, cv2.CC_STAT_HEIGHT]
width = stats[i, cv2.CC_STAT_WIDTH]
if left == 0 or top == 0 or left + width >= cols or top + height >= rows: # touching edge, skip
continue
char_img = np.ndarray([height, width], np.uint8)
char_img.fill(max_value) # set to white background
for row in range(top, top+height):
for col in range(left, left+width):
if labels[row, col] == i or labels[row, col] == 0: # include label = 0 to obtain background values
char_img[row-top, col-left] = img[row, col]
output_path = output_folder + str(17)+"_"+ str(j).zfill(4) + ".bmp"
#output_path2 = "C:/xampp/htdocs/DeepOCR/" + name +"/" + str(k)+"_"+ str(j).zfill(4) + ".bmp"
j = j + 1
cv2.imwrite(output_path, char_img)
#cv2.imwrite(output_path2, char_img)
Create Features for new data¶
In [4]:
folder_path = os.path.join(os.getcwd(), name+"/"+name)
#folder_path = "C:/xampp/htdocs/DeepOCR/"+name+"/"+name
file_names = next(os.walk(folder_path))
#print(file_names[2])
if '.bmp' in file_names[2]:
print(len(file_names[2]))
data_set = []
count_sample = 0
for i in range(len(file_names[2])):
image_path = file_names[0] + "/" + file_names[2][i]
#print(full_path)
img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
img = cv2.resize(img, (28, 28))
#plt.imshow(img, 'gray')
#plt.show()
features = "|features"
for row in range(len(img)):
for col in range(len(img)):
features = features + " " +str(img[row, col])
#print(features)
#return features
data_line = features + "\n"
data_set.append(data_line)
count_sample += 1
print(" done with " + str(count_sample) + " samples")
with open("dataset_"+name+".txt", "w") as outfile:
for data_line in data_set:
outfile.write(data_line)
Model¶
In [5]:
np.random.seed(0) #กำหนดจุดกำเนิดเลขสุ่มแบบตายตัวถ้าอยากให้การทดลองถูกทำซ้ำ
input_dim_model = (1,28,28) # images are 28 x 28 with 1 channel of color (gray)
input_dim = 784
num_output_classes = 286
import cntk
input = cntk.input_variable(input_dim_model)
label = cntk.input_variable(num_output_classes)
#เตรียมพาธของข้อมูลฝึกและข้อมูลทดสอบ
data_found = False
for data_dir in ["."]:
train_file = os.path.join(data_dir, "rantrain_AllOverlap_035-320.txt")
test_file = os.path.join(data_dir, "dataset_"+name+".txt")
if os.path.isfile(train_file) and os.path.isfile(test_file):
data_found = True
break
if not data_found:
raise ValueError("Your data files are not available. Please check it out if you put them in the same fol")
print("Data directory is {0}".format(data_dir))
print("Train-data path is " + train_file)
print("Test-data path is " + test_file)
def create_model(features):
with cntk.layers.default_options(init=cntk.glorot_uniform(), activation=cntk.relu):
h = features
h = cntk.layers.Convolution2D(filter_shape=(5,5),
num_filters=32,
strides=(1,1),
pad=True, name='first_conv')(h)
h = cntk.layers.MaxPooling(filter_shape=(3,3),
strides=(2,2), name="first_max")(h)
h = cntk.layers.Convolution2D(filter_shape=(3,3),
num_filters=48,
strides=(1,1),
pad=False, name='second_conv')(h)
h = cntk.layers.MaxPooling(filter_shape=(3,3),
strides=(2,2), name="second_max")(h)
h = cntk.layers.Convolution2D(filter_shape=(3,3),
num_filters=64,
strides=(1,1),
pad=False, name='third_conv')(h)
h = cntk.layers.Dense(96, activation=cntk.relu, name='classify')(h)
h = cntk.layers.Dropout(dropout_rate=0.5)(h)
r = cntk.layers.Embedding(num_output_classes)(h)
return r
In [7]:
def create_reader(path, is_training, input_dim, num_label_classes):
featureStream = cntk.io.StreamDef(field='features', shape=input_dim, is_sparse=False) #784
deserailizer = cntk.io.CTFDeserializer(path, cntk.io.StreamDefs(features = featureStream))
return cntk.io.MinibatchSource(deserailizer,
randomize = is_training, max_sweeps = cntk.io.INFINITELY_REPEAT if is_training else 1)
def create_criterion_function(model, labels):
loss = cntk.cross_entropy_with_softmax(model, labels)
errs = cntk.classification_error(model, labels)
return loss, errs # (model, labels) -> (loss, error metric)
def print_training_progress(trainer, mb, frequency, verbose=1):
training_loss = "NA"
eval_error = "NA"
if mb%frequency == 0:
training_loss = trainer.previous_minibatch_loss_average
eval_error = trainer.previous_minibatch_evaluation_average
if verbose:
print ("Minibatch: {0}, Loss: {1:.4f}, Error: {2:.2f}%".format(mb, training_loss, eval_error*100))
return mb, training_loss, eval_error
z = create_model(input/255.0)
loss, label_error = create_criterion_function(z, label)
learning_rate = 0.3
lr_schedule = cntk.learning_rate_schedule(learning_rate, cntk.UnitType.minibatch)
learner = cntk.sgd(z.parameters, lr_schedule)
trainer = cntk.Trainer(z, (loss, label_error), [learner])
def evaluate(reader):
trainer.restore_from_checkpoint("Model_35_320-1.1.dnn")
test_input_map = {
input : reader_test.streams.features,
}
reader_test = create_reader(test_file, False, input_dim, num_output_classes)
evaluate(reader_test)
predictor = cntk.softmax(z)
# Read the data for evaluation
reader_eval = create_reader(test_file, False, input_dim, num_output_classes)
eval_minibatch_size = count_sample
eval_input_map = {input: reader_eval.streams.features}
data = reader_eval.next_minibatch(eval_minibatch_size, input_map=eval_input_map)
#img_label = data[label].asarray()
img_data = data[input].asarray()
# reshape img_data to: M x 1 x 28 x 28 to be compatible with model
img_data = np.reshape(img_data, (eval_minibatch_size, 1, 28, 28))
predicted_label_prob = [predictor.eval(img_data[i]) for i in range(len(img_data))]
#gtlabel = [np.argmax(img_label[i]) for i in range(len(img_label))]
pred = [np.argmax(predicted_label_prob[i]) for i in range(len(predicted_label_prob))]
#print("Label :", gtlabel[:30])
#print("Predicted:", pred)
In [8]:
consonant = ['#','$',' ','&','(single quote)','(',')','(ดอกจันทร์)','+',',','-(ลบ)','.(จุด)','(slash)','0','1','2','3','4','5','6','7','8','9',' ',' ','<',' ','>','?','@','A','B','C','D'
,'E','F','G','H','I','J','K','L','M','N','O','P','Q','R','S','T','U','V','W','X','Y','Z','[','(back slash)',']',' ',' ',' ','a','b','c','d'
,'e','f','g','h','i','j','k','l(แอลเล็ก)','m','n','o','p','q','r','s','t','u','v','w','x','y',' ','{','(ขีดตั้ง)','}',' ',' ',' ',' ',' ',' '
,' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ','ก', 'ข', 'ฃ', 'ค'
,'ฅ', 'ฆ', 'ง', 'จ', 'ฉ', 'ช', 'ซ', 'ฌ', 'ญ', 'ฎ', 'ฏ', 'ฐ', 'ฑ', 'ฒ', 'ณ', 'ด', 'ต', 'ถ', 'ท', 'ธ', 'น', 'บ'
, 'ป', 'ผ', 'ฝ', 'พ', 'ฟ', 'ภ', 'ม', 'ย', 'ร','ฤ(ตัวรึ)', 'ล', 'ฦ(ตัวลึ)', 'ว', 'ศ', 'ษ', 'ส', 'ห', 'ฬ', 'อ', 'ฮ','ฯ',' ','อั(ไม้หันอากาศ)'
,'า',' ','อิ(สระอิ)','อี(สระอี)','อึ(สระอึ)','อื(สระอือ)','อุ(สระอุ)','อู(สระอู)',' ',' ',' ',' ',' ','฿','เ',' ','โ','ใ','ไ','ๅ(ลากข้าง)','ๆ','อ็','อ่(ไม้เอก)','อ้(ไม้โท)',
'อ๊(ไม้ตรี)','อ๋(ไม้จัตวา)','อ์(การันต์)','อำ',' ',' ','๐(ศูนย์ไทย)','๑','๒','๓','๔'
,'๕','๖','๗','๘','๙',' ',' ',' ',' ',' ',' ','ญ(ติด)','ฐ(ติด)','สั(ติด)','ป็(ติด)','ปั(ติด)',' ป้(ติด)','ฟั(ติด)'
,'ฟ้(ติด)','ฝั(ติด)','ฝ้(ติด)','อี่(ติด)','อี้(ติด)','อื่(ติด)','อื้(ติด)','อึ่(ติด)','อึ้(ติด)','อ้โ(ติด)','อ้ใ(ติด)','อ้ไ(ติด)'
,'อ์โ(ติด)','อ์ใ(ติด)','อ์ไ(ติด)','อ็ไ(ติด)','สิ(ติด)','สี(ติด)','สื(ติด)','ส้(ติด)','ศื(ติด)','ศี(ติด)','ศั(ติด)','ชี(ติด)','ชื(ติด)','ชั(ติด)','ช็(ติด)',
'ชโ(ติด)','ชไ(ติด)','ชใ(ติด)','ซี(ติด)','ซื(ติด)','ซึ(ติด)','ปิ(ติด)','ปี(ติด)','ปื(ติด)','ปโ(ติด)','ปไ(ติด)','ปใ(ติด)','ฝี(ติด)','ฝึ(ติด)','ฝื(ติด)',
'ฝใ(ติด)','ฟิ(ติด)','ฟื(ติด)','ฟโ(ติด)','ฟไ(ติด)','ฟใ(ติด)','อีโ(ติด)','อีไ(ติด)','อีใ(ติด)','อ่โ(ติด)','อ่ไ(ติด)','อ่ใ(ติด)','ศิ(ติด)','ศึ(ติด)',
'สึ(ติด)','สั(ติด)']
count_arr = 0
sum_arr = 0
row = 0
col = 0
data_set = []
i = 0
for i in range(len(pred)):
plt.imshow(img_data[i].reshape(28,28), cmap="gray_r")
plt.axis('off')
plt.show()
#folder_path = "C:/xampp/htdocs/DeepOCR/"+name
file_names = next(os.walk(folder_path))
image_path = file_names[0] + "/" + file_names[2][i]
#index = image_path.find('DeepOCR')
#str_path ="http://localhost/"+image_path[index:]
print("ตำแหน่ง : ",pred[i])
#worksheet.write(row, col, pred[i])
for cons in range (0,286):
if(cons == pred[i]):
c = pred[i]
print("ทำนาย : ", consonant[c])
#worksheet.write(row, col + 1, consonant[c])
single_sample_prob = [predictor.eval(img_data[i])]
sum_single = sum(single_sample_prob[0][0])
max_eval = np.max(single_sample_prob)
loss = -math.log(max_eval/sum_single)
per = format((max_eval * 100),'.2f')
print("ค่าความมั่นใจ :" + " "+str(per)+"%")
#worksheet.write(row, col + 2, str(per))
#insert_data(name,pred[i],consonant[c],str(per)+"%",str_path)
row = row + 1
#workbook.close()
Classification¶
In [9]:
j = 0
k = 1
base_folder = "D:/cntk1.1/Examples/Image/DataSets/MNIST/OCR-Database/"+name #เปลี่ยนชื่อเอา
folder_path = base_folder + "/" +name
file_names = next(os.walk(folder_path))
print(len(file_names[2]))
for i in range(len(pred)):
image_path = file_names[0] + "/" + file_names[2][i]
#print(image_path)
for cons in range (0,286):
if(cons == pred[i]):
c = pred[i]+35
img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
output_classify_folder = base_folder + "/classify_"+ str(c) + "/"
try:
os.mkdir(output_classify_folder)
except OSError as exc:
print("folder " + output_classify_folder + " exists")
output_classify = output_classify_folder + str(k) + ".bmp"
print(output_classify)
j = j+1
k = k+1
cv2.imwrite(output_classify,img)
#print(j)
In [ ]: